1. Field of Disclosure
Aspects of the present invention relate to decision support systems, and more particularly, to decision support systems that employ hybrid rules-based and case-based reasoning.
2. Discussion of Related Art
Decision support systems (DSSs) are typically designed to provide users with near real-time access to complex bodies of knowledge. Often the users of DSSs are themselves experts in the bodies of knowledge at which the DSS is targeted, and these users harness the DSS to better apply their knowledge to a particular set of facts. Thus, conventional DSSs accept a set of facts and, based on these facts and the content of the pertinent body of knowledge, provide potential conclusions drawn from application of the pertinent body of knowledge.
A DSS may be broadly categorized according to how the DSS stores and applies, or reasons with, a body of knowledge. Conventionally, there are at least three such categorizes of DSS reasoning. These categories include rule-based reasoning (RBR), case-based reasoning (CBR) and hybrid RBR/CBR. RBR uses facts derived from a well-constructed domain theory to produce knowledge. In rules-based systems, rules, stored in the form of logical implications, or if/then statements, form the building blocks of the decision making process.
CBR recognizes the power of individual solutions and solves problems by retrieving cases from a database of prior cases and adapting the solutions of the retrieved cases to fit the use case data. In CBR, instead of using general domain knowledge to solve a current case, specific knowledge from previous cases is used. Conventionally, CBR systems share the following cycle:    1. RETRIEVE. The most similar case to the current one is retrieved. In some CBR systems, syntactical similarities are used and in others semantic similarities are used. A commonly used method for case retrieval is neighborhood matching where the case similarity may be computed as shown below, where wi are feature weights, Fi_i is the ith feature in the input case, and Fi_o is the ith features in the retrieved case:
            ∑              i        =        1            n        ⁢                  ⁢          wiXsim      ⁡              (                  Fi_i          ,          Fi_o                )                        ∑              i        =        1            n        ⁢                  ⁢    wi      2. REUSE. The information in the most similar case is reused to solve the current case. This reuse may include simply applying the solution of an old case the new case. Reuse may also include adapting, by either reusing the past case solution after a transformation that matches to the current case, or by applying a method used in the solution of the past case.    3. REVISE. The solution is evaluated in the context of the current case and revised appropriately. Often this evaluation is conducted by an expert and may last from a few hours to a few months.    4. RETAIN: The case base is updated with the solution of the current case. In case of success, the case base is updated with the current case and, optionally, index strengths that contributed to the successful solution are strengthened. In case of failure, index strengths that contributed to the failure are weakened.
Conventionally, cases are represented using a variety of data structures. One popular technique is the dynamic memory model, where cases with similar structure are organized in a general structure, known as a generalized episode (GE). A GE contains norms, which are features common to all cases, and indices, which are features that discriminated between cases. Another popular technique is the category and exemplar model, which organizes the cases in a network of categories, semantic relationships, cases, and index pointers.
Conventional case indexing schemes include case indexing by the features with the most predictive power, difference-based indexing and explanation-based generalization methods. When indexing by the features with the most predictive power, features that are responsible for providing a solution to the case are indexed. Difference-based indexing uses indices that differentiate cases from other similar cases. Explanation-based generalization methods form an abstract case from common features and the abstract case is indexed via these common features.
A number of CBR systems have been developed in both industry and academia. These CBR systems include CBR Express/ART-Enterprise from Inference Corporation, which are used primarily for help desk applications, Remind from Cognitive Systems Inc., and ReCall from ISoft. Non-commercial CBR systems for specific applications, such as clinical audiology, heart failure and meal planning, have also been developed.
Hybrid RBR/CBR systems developed in response to the deficiencies of RBR and CBR, when each is applied individually. Systems using this reasoning approach typically marshal reasoning activates between RBR and CBR engines in an attempt to reap the benefits of both.